Improving Classification Accuracy Using Combined Filter+Wrapper Feature Selection Technique

Large datasets such as medical datasets contain a lot of redundant or irrelevant attributes which will degrade the performance of the classifier. Feature selection methods help to improve the classification accuracy by providing relevant feature subset that best describes the dataset. The generated feature subset will be tendentious to the feature selection approach and the classification accuracy will rely on the relevancy of the features selected. Hence the process of deciding appropriate feature selection algorithm has a crucial role in providing relevant features. Traditional techniques for feature selection are filter methods and wrapper methods. In this paper, a novel hybrid feature selection approach that efficiently incorporates the benefits of both filter and wrapper methods is proposed. The features were first ranked based on the ranking criteria's and then a wrapper algorithm is invoked to generate a subset from the ranked features. A minimal redundancy maximal relevant feature subset is generated through the proposed method.